Information Need Based Answer Summarization For Community Question Answering

In recent years, more and more community question answering portalsemerge as a knowledge sharing platforms, and provide new opportunities for QAsystem. They offer plenty of questions and corresponding answers information toQA system. They can well meet the users’ individualized information demandbecause of its interactivity and the characteristics of open. There are lots ofonline community question answering portals including Baidu zhidao, Sousouwenwen and Tianya wenda appear. It illustrates that the traditional search enginebased on key words query has been unable to allow the users to search their ownpersonalized information needs quickly and accurately. The community questionanswering portals allow users to ask for information by post questions. Thequestions may contain the content of questions and question description forcontext complement. All the users can answer the questions, and the questionerwould choose one of the answers for best answer.However, when the users are not familiar with the field of information theyneed, they may not be able to organize their own query language to catch theirinformation needs. This paper expands the users’ query on community questionanswering portals, context information and answers are also extended at the sametime. We can get all aspects of information needs questions related to the users’query by co-clustering questions set contain different information needs and theiranswers set.For a same question under different description (context information), theanswer may be completely different. For example, the question "how to buy amobile phone?".When the question is described,"where to buy" and "how to b uya cheap one", the information needs users want to get is totally different. In orderto resolve this problem, introduces the constraint conditions into theco-clustering model based on context information.Meanwhile, the answers provided by users are quality uneven. Sometimesthe useful information may be attached to the useless information even falseinformation at the same time. So the high redundancy of the information isdifficult to be applied to question-answering system. In this paper, we collectedtextual features and non-textual features of the answers to establish the answersorting model. Summarize the answers from ranked information in each cluster.A large number of answers must contain much duplicate information, andthe answers may have a difference in the representation. It will be difficult todetect by calculated the similarity. In this paper, we establish an answer similarity detection model, through leveraging multiple classifiers voting method,detecting the duplicate answers, and remove them.